willingness to pay for locally produced foods: a customer...
Post on 01-Apr-2018
222 Views
Preview:
TRANSCRIPT
Willingness to pay for locally produced foods: A customer intercept study of direct market and grocery store shoppers
Kim Darby, Marvin T. Batte, Stan Ernst and Brian Roe Department of Agricultural, Environmental and Development Economics,
The Ohio State University
Corresponding Author: Marvin T. Batte
Agricultural, Environmental, and Development Economics The Ohio State University
2120 Fyffe Rd. Columbus, OH 43210-1067
PH: 614-292-6406 Email: batte.1@osu.edu
Selected Paper prepared for presentation at the American Agricultural Economics Association Annual Meeting, Long Beach, California, July 23-26, 2006
Copyright 2006 by Kim Darby, Marvin T. Batte, Stan Ernst and Brian Roe. All rights reserved. Readers may make verbatim copies of this document for non-commercial purposes by any means, provided that this copyright notice appears on all such copies.
1
Locally-grown produce has an appeal that draws many to roadside stands each summer.
Likewise, we are seeing more grocery stores that feature foods that are identified as locally
grown or processed, suggesting that stores have assessed the market for these differentiated
products to be sufficient to warrant increased shelf space. Freshness and taste are obvious
reasons for consumer preference for these goods, but also important may be home-bias, which
often is attributed to ethnocentrism, or less pejoratively, “hometown pride” (Scarpa 2005).
Home-bias can refer to local consumption as a means to support local business. Alternatively,
home-bias may be interpreted as an attempt to preserve a certain way of life, a countryside
without large mono-crop cultivation and urban sprawl. Moreover, some consumers may
associate local food systems with the concept of “sustainability,” referring to a local food system
which uses less energy in its supply chain, which supports a regional economy, and which is less
dependent on trade. Whatever the motive, there is substantial evidence that some consumers are
willing to pay premium prices for food characterized as locally produced.
Marketing differentiated food products locally provides an opportunity for farms to capture
a greater share of consumers' food budgets, and for rural communities to generate greater
incomes. Successful product differentiation and profitable product placement require more
specialized knowledge of those food characteristics valued by consumers. As such, clarifying
and quantifying the appeal of locally-grown produce provides valuable information to those
interested in marketing. Where other studies have clarified the concept of local into levels of
distance, this study additionally attempts to decompose the type of quality communicated by the
“locally-grown” signal. Moreover, where other studies have used qualitative methods to assess
preferences, the quantitative methods used in this study provide information useful for pricing
decisions for both direct markets and retail grocery.
2
The objective of this study is to explore the potential for differentiation in fresh produce
through estimation of willingness to pay (WTP) for selected characteristics. Specifically, this
study relies on a customer-intercept survey and a choice experiment of food shoppers in a variety
of direct markets and traditional grocery stores. Conjoint analysis is used to evaluate WTP for
characteristics related to locally grown fresh strawberries.
Literature Review
In the past 15 years consumer demand for niche products (including organic, natural, and
locally grown) has grown substantially (Dimitri and Greene). Consumers value locally produced
foods or foods produced with a particular technology because they perceive the products to be
healthier, to be more environmentally friendly, or to be more supportive of small scale
agriculture and local rural communities. This preference may translate to a willingness to pay a
premium price for that product.
Govindasamy and Italia (1999) showed that younger consumers, regardless of gender, paid
higher premiums for organic produce, as did females with higher annual incomes. They also
found that the probability of paying a premium goes down as the number of individuals in the
household rises. Thompson and Kidwell (1998) found that families with children were more
likely to buy organic produce than those without children, whereas Loureiro and Hine (2002) and
Wang and Sun (2003) show the opposite. Moreover, Williams and Hammitt (2000, 2001) found
few distinctions between organic and conventional fresh produce consumers. Batte et al., in a
study of multi-ingredient organic foods, found that shoppers who were aware of the National
Organic Program organic seal for food products were more likely to be willing to pay a premium
price. Conditioned on a willingness to pay a premium, they found older consumers and those
with higher income per household member were willing to pay higher premia. Females,
3
consumers with children, and consumers who were intercepted in a whole foods health store
were also willing to pay prices that were significantly larger than their counterparts.
Consumer acceptance and willingness to pay for genetically modified food has been
studied with stated choice methods using contingent valuation (Ganiere 2004), experimental
auctions (Huffman 2003) and observed choice (James 2005). Onyango (2004) used conjoint
methods to show that WTP estimates for GM bananas were positive for “additional oxidants”,
followed in magnitude by “less pesticide-use” and “ripens longer.” The study showed negative
WTP for the actual use of GM technology, indicating that consumers would need to be
compensated to purchase a GM product.
In addition to consumer willingness to pay for food attributes perceived to relate to food
safety and nutrition, consumers have been shown to be willing to pay for ideological functions of
food production systems. Williams and Hammitt (2000, 2001) and Underhill and Figueroa
(1996) show that WTP to pay for organic foods is related to a perception of its being more
environmentally friendly and supportive of small-scale agriculture and local rural communities.
Choice experiments have been used to show positive WTP for eco-labels on seafood (indicating
sustainable fishery management) certified by an independent third party (Wessells et al.1999,
Johnston et al. 2001, and Teisl et al. 2002). Aquino and Falk (2001) and Hurley and Kliebenstein
(1998) show that, respectively, 87.5% and 67% of respondents (self-selected environmentalists)
agreed with the statement that they would pay a price premium for “wolf-friendly” beef or beef
raised with production methods that improve water quality and animal welfare. Contingent
valuation of eco-labels for fresh apples also showed a price premium of between $0.10 and $0.50
(Loureiro and McCluskey 2002).
4
Whether for safety, quality, or an origin-based ideology, country of origin (CoOL) and
region of origin (ROO) labeling have been shown to garner price premiums. In an experimental
auction, Umberger (2003) showed that the majority of consumers (69%) had positive WTP
estimates for CoOL certified beef. However, in a related study, Louriero and Umberger (2004)
use conjoint analysis to estimate WTP for CoOL vs. other safety labels on beef. They show that
CoOL labeling does not have positive WTP estimates, while food safety certification and
traceability certification do. They also show that age and gender are significant determinants of
WTP for food safety, while income is a significant determinant of WTP for traceability. Mabiso
(2005) used an auction experiment to show that the CoOL label garnered premiums for fresh
produce. In the case of apples, 79% paid a premium, $0.48 on average, and in the case of
tomatoes, the figures were 72% and $0.49, respectively.
On the regional level, both generic ROO labels and protected designation of origin (PDO)
or protected geographical identification (PGI) labels garner premiums. Scarpa, et. al. (2005) used
conjoint analysis to estimate WTP for the PDO label on olive oil as well as a CoOL label on
oranges and table grapes. In the case of table grapes, they found WTP for these attributes to
have about the same impact on preferences as more common characteristics such as color and
ripeness. The CoOL label on oranges had nearly four times the impact than that of firmness and
absence of imperfections. Loureiro and McCluskey (2000) cite cultural identification as well as
perceived quality to account for premiums found using a hedonic model on PGI Galecian veal.
Van der Lans et. al. (2001) isolates two reasons for WTP for ROO on olive oil: the direct effect
(i.e. cultural identification) and its indirect effect through perceived quality. Using conjoint
techniques, the study shows that, while PDO labels only have the indirect effect, ROO labels
have the additional direct effect, especially for consumers from the same region.
5
Buchardi et al (2005) used contingent valuation to estimate WTP for milk from the
consumer’s own region. They found that “consumers perceive fresh milk from local farmers as a
trustful, high quality product, and that consumers are interested in supporting local producers.”
(pg. 2). However, in two separate experiments, one that is incentive compatible and the other
purely hypothetical, WTP estimates are significantly different, indicating a hypothetical bias.
Campbell et al. (2004) used conjoint analysis to estimate part worth values for Satsuma
Mandarins, including physical attributes (skin color, size, blemishes, and seeds), an own-region
cue (in-state vs. US) and an organic cue. Their results suggested the location attribute was the
least important product attribute, and the physical attribute “seedless” had the greatest impact on
likelihood of purchase, even greater than price.
Brown (2003) found that in response to the question, “Would you pay a price that was
lower, the same, or higher for products labeled ‘locally grown’ versus unlabeled products of the
same quality?,” 58% of food consumers were unwilling to pay a premium for food products
labeled as “locally grown” provided that the unlabeled products were of the same quality,
whereas 16%, 5% and 1% said they would pay more than a five, ten or 25 percent premia,
respectively. Socioeconomic characteristics that significantly influenced WTP were gender
(females were more likely to pay a higher price). Survey respondents with annual household
income of $50,000 or more or with a graduate or professional degree were more likely to pay a
higher price for local food.
Data and Methods
During the period August 2005 - January 2006, researchers interviewed shoppers at 17
locations. These included six farm markets, four farmers' markets, and seven retail grocery
6
stores.1 Two of the grocery stores were located in small towns. Two were in urban locations
within Columbus Ohio, and three were in suburban locations surrounding Columbus.
Interviewers randomly selected consumers, asked for their participation, and verified that each
participant was at least 18 years of age. After completing a series of eight choice experiments,
the customers completed a survey that included attitudinal questions and elicited economic and
demographic information for the respondent. Table 1 depicts the makeup of the sample by
market type and location.
Table 1. Number of Respondents and Choice Experiment Observations by Intercept Location
Market Typea Respondents Choice Experiment
Observations Direct Markets Farmer's Markets Metro 32 256 Rural 96 768 Farm Markets Metro 105 840 Rural 34 272 Direct Market Sub-Total 267 2136 Grocery Stores Rural 80 640 City Center 78 624 Suburban 105 840 Grocery Store Sub-Total 263 2104 Total 530 4240 a Rural refers to areas that are outside a commuting distance of a major urban center. Metropolitan area refers to locations that are just beyond the suburban rim of an urban center, but still within commuting distance. City center refers to locations in the inner city of the Columbus metropolitan area, and suburban refers to location in bedroom communities surrounding Columbus.
Choice Analysis and Willingness to Pay Measures
Willingness-to-pay measurements are grounded in utility theory. Hanemann (1991)
outlines the theoretical underpinnings as a utility maximization problem subject to a budget 1 Farmers’ markets refer to open-air markets with multiple stalls representing a variety of farm and/or marketing operations. Farm markets refer to either indoor or outdoor venues where a single farm operation is selling goods.
7
constraint. The consumer chooses the level of the good X that maximizes utility, producing the
traditional Marshallian demand curve X(p, y, q), where p is market price, y is income and q, is
the quality of the good, fixed exogenously. The resulting indirect utility function is V(p, y, q).
Identifying a change in a good’s quality from q0 to q1, the measurement of value is
V(p, y-WTP, q1) = V(p, y, q0). (1)
where WTP is the amount the consumer would be willing to pay for the improved quality,
maintaining constant utility.
The estimate is shown more directly using the dual problem: expenditure minimization
constrained by a given utility level (Lusk 2004). The dual produces the Hicksian demand curve
X(p, U, q) and indirect expenditure function M(p, U, q) so that
WTP = M(p, U, q0) - M(p, U, q1). (2)
where U is a constant utility level.
The random utility model (RUM) is used to analyze choices and estimate WTP. When
respondents are asked to make a choice of whether or not to pay a given dollar amount, a
positive response is interpreted as their WTP. In order to find the central tendency of WTP from
a sample, the positive responses are fitted to a probability function. This is modeled as the
probability that the utility derived from the good associated with that choice is greater than the
alternative,
P[(V1 + �1) > (V0 + �0)] (3)
or P[(�1 – �0) < (V1 – V0)] (4)
where V is the indirect utility and � is the error term.
8
Estimates of WTP are based on the mid-point of this function, i.e. the point at which the
probability of a positive response is .5. The most basic version of this model includes only the
socioeconomic attributes of respondents as variables.
P[{�1(s) - �2(s)} < (V1(s)- V2(s)}] (5)
This is the model used in contingent valuation choice analysis. Conjoint analysis offers a more
sophisticated alternative that allows the quality of a good to be decomposed and their
contribution to total WTP estimated.
Lancaster (1966) builds the conceptual framework for conjoint analysis by clarifying that
utility is gained from the characteristics of a good rather than the good itself. Characteristics are
objective qualities of a good while attributes, on the other hand, are what the characteristics
represent, and are the real source of an individual’s utility. For example, characteristics of fresh
strawberries may include color, size, absence of blemishes and type of packaging. These
characteristics represent attributes, such as taste, convenience, and nutrition. Lancaster
maintained that utility should be a function of characteristics rather than attributes as the former
are measurable, so that,
u = U(t1, t2, t3) (6)
t = BX (7)
where t represents the characteristics of a good X, based on the consumption technology
matrix B (i.e. the amounts of each characteristic which are predetermined in each good).
Louviere, Hensher and Swait (2000) further refine these distinctions, making utility a
function of consumption services. Each characteristic of a good is associated with a consumption
service which it provides. Moreover, they suggest that utility maximization is based on expected
services of a good because consumers do not have complete information. In practice, the model
9
forces a one-to-one correlation between services and characteristics. However, these theoretical
distinctions are helpful in conceptualizing characteristics as signals that communicate value.
Analysts can then examine which signals most effectively communicate the service that
ultimately provides utility.
Conjoint analysis proceeds with a good that is decomposed according to the Lancastrian
model and measures its part-worth values. It uses a more sophisticated RUM to estimate the
probability that utility of the ith individual derived from the jth alternative is greater than other
alternatives in the set,
Probij = P[(Vij + �ij) > (Via + �ia ;a =1,2,…J, a � j)] (8)
or P[(�ij – �ia) < (Vij – Via)] (9)
where Vij denotes the individual’s indirect utility from choosing product j and �ij is an error term.
Results
Descriptive statistics for the sample are reported in Table 2. Mean age of consumers
interviewed was 49.5 years. Seventy-two percent were female. The average household size was
2.6 people, and just over 33 percent of households had children aged 19 or younger. Our
respondents were more highly educated than average Ohioans – 78 percent had some education
post high school. Mean annual household income was $81,891, and was about $12,000 more for
consumers intercepted in grocery stores than for those contacted at direct markets.
10
Table 2 Consumer sample demographic and economic variables Mean or Percent
Variable Full
Sample Direct Grocery Number of respondents 530 267 263 Mean Household Income ($) 81,891 75,768 87,963 *** Household weekly expenditure for food ($) 101 96 106 *** Household weekly expenditure for produce ($) 28 28 28 Mean age of respondent (years) 49.5 49.2 49.8 Gender (Percent female) 72.0 69.4 74.7 *** Number in household: 2.6 2.6 2.7 ** Percent households with children 33.4 34.2 32.5 Percent with post-high school education 78.1 78.9 77.3 Percent white 92.1 89.8 94.5 *** Percent metropolitana 61.1 55.1 67.1 *** Percent U.S. citizens 95.3 94.9 95.7 * One, two and three asterisks indicates a difference of the means for the direct and grocery shopper groups that was significant at the 0.10, 0.05 or 0.01 probability levels, respectively. a Percent of respondents who indicated that they lived in a city or surburb of a city.
Respondents were asked to identify their reasons for purchasing locally grown foods
from a list of options. The list included freshness, taste, nutrition, safety, to support the regional
economy, to support local businesses, and to maintain a direct connection with the source of
food. They were also asked to rank their top three choices from among this list. Table 3 reports
the overall percentage of respondents who identified a reason as one of their top three choices.
Freshness was the most frequently cited among both consumer groups, although more so for the
grocery store customers. Supporting local businesses was the next most frequently cited reason
for both groups, although this reason was cited about 20% less often than freshness. Taste also
ranked high at 63% of all consumers, but was higher still (69.5%) for direct market customers.
11
Table 3 Primary reasons to purchase locally grown foods Percent Citing as Most Importanta Full Direct Grocery Freshness 91.3 89.0 93.8 Support local businesses 71.0 70.4 71.6 Taste 62.7 69.5 55.5 For direct connection with source of food 20.7 22.8 18.6 Nutrition 21.5 21.7 21.4 Support regional economy 20.8 16.6 25.2 Safety 11.9 10.0 14.0 a Percent of respondents citing this reason as one of the three most important motives for purchasing local foods.
Respondents were also asked about their reasons for selecting one food product relative to
others when shopping at a grocery store. The list of selection reasons and mean evaluation
scores for each are presented in Table 4. The top ranked reason was taste/quality (3.54), which
was not based on labels but on past experience. Price was also an important factor (2.9),
followed by ease of preparation (2.62). Following this were several health-related food
characteristics, including low in transfat, fat, cholesterol, and sodium. Product brand ranked in
the lower half of the list of reasons to purchase.
The rankings were similar across the sub-sample groups based on store type, but there
differences between these groups. Customers intercepted at direct markets gave higher
importance scores to food taste/quality and foods labeled as natural or organic. Grocery store
shoppers gave higher mean importance scores to health related measures, ease of preparation,
convenience of packaging, and product brand. Interestingly, there was no significant difference
in the importance given to product price for the two groups.
12
Table 4. Characteristics rated as important in food purchase decisions.a Total Store Type Means Direct Grocery Taste/quality (from past experience) 3.54 3.57 3.52 Price 2.90 2.91 2.88 Ease of preparation 2.62 2.50 2.74 Low trans-fats 2.52 2.48 2.57 Low-cholesterol 2.35 2.27 2.44 Low-fat 2.23 2.12 2.34 Low-sodium 2.22 2.11 2.34 Labeled as natural 2.18 2.23 2.13 Brand 2.18 2.14 2.22 Convenience of packaging 2.10 2.04 2.18 Low calorie 2.08 1.96 2.20 Labeled as organic 2.02 2.14 1.90 Labeled as Heart-Smart 1.96 1.92 2.00 a The question read "In general, when you purchase food of any type at the grocery store, how would you rate the importance of the following characteristics in your decision? ". Responses are coded: Not important=1, Somewhat Important=2, Important=3, very important=4. Bold indicates significant at least at the .05 level
Conjoint Study
The design used in this study includes four variables: price, location of production,
size/type of producing firm, and whether or not there is a product freshness guarantee. All other
characteristics are assumed equal.
The location-of-production cues provide the most general information on the value of
origin and degree of localness. The options for location are “grown on a nearby farm,” “grown
in Ohio,” “grown in the U.S.,” and unidentified, which is displayed as a blank. The size of the
producing firm cue is intended to capture the element of “local” that implies a social ideology
around small-scale agriculture and supporting locally owned businesses. The name of producer is
displayed as either “Fred’s Berry Farm” or “Berries Inc.,” implying small family farm versus a
13
large corporate farm, although these cues will not be interpreted in the same way by each
respondent. To capture the element of local that is exclusively related to its perceived freshness,
we included a freshness guarantee cue. The label either indicates “harvested yesterday” or was
left blank. Respondents are not given any information regarding the type of certification process
involved with the guarantee. Lastly, the price variable included five levels, ranging from $2.00
to $4.00 in $.50 increments.
We implemented our study using a full factorial experimental design. There are
2*2*4*5=80 possible product profiles. We implemented this design by randomly sampling from
these profiles for each product. Thus, each respondent viewed different product comparisons
based entirely on chance. In the event that two identical products were randomly selected for
comparison, a new drawing was made automatically. Also, we were willing to assume that all
consumers preferred a product with a freshness guarantee to none, if all other attributes were
identical, and prefer less costly products to more costly ones, again assuming all other product
attributes are identical. When two products were randomly selected and one product dominated
the other as described in the previous sentence, the computer automatically drew another product
pair.
The conjoint experiment was administered in face-to-face interviews. Customers were
approached randomly and told that the interviewer was a student in the College of Agriculture at
The Ohio State University working on a research project. They were also told that the survey,
which regarded their preferences for fresh produce, would take approximately 7-8 minutes to
complete. After ascertaining the respondents were over the age of 18, they sat at a table in the
produce section and looked at the screen of a laptop computer. They were then given the
following scenario:
14
We will begin by asking you to compare several hypothetical fresh strawberry
products. Imagine that you are out for a drive in the country and decide to stop at a
roadside produce market. There you see two fresh strawberry displays. The
strawberries are being sold in one-quart containers. In the following slides you will
see the product information presented in these displays. Other than the information
that is provided, assume that the berries are identical in terms of size, color, ripeness,
etc. Of course, any purchase that you make will reduce the funds that you and your
family have available for other purchases.
The final comment is intended as a budget-constraint reminder to reduce hypothetical
bias (Loomis 1997). Next, the interviewer generated the first paired comparison and gave a brief
overview of the options to help the respondent process the information. The following is an
example:
Figure 1 Display used for conjoint experiment
Product 1: Product 2: Where Produced: Grown in the U.S. Grown nearby Producer: Berries, Inc. Fred's Berry Farm Freshness guarantee? Harvested Yesterday Price: $ 2.00 $ 3.00
The respondent was given the choice to indicate a preference for either product or to indicate no
preference. This process was repeated eight times for each respondent.
15
In our study, the levels of locality were not conducive to a metric scale, so we chose to
use of the part worth utility model. In addition, the other two attributes (farm size and freshness)
were originally conceptualized as dummy variables. Only with the price variable, which lends
itself readily to a metric scale, do we assume a continuous negative relationship, so that
ijjjij pxV επβ ++= (10)
where Vij denotes the individual’s indirect utility from choosing product j; xj is a vector of
product attribute level j’s; pj is price for product j; � is a conformable vector of coefficients and
� is a conformable coefficient to be estimated; and �ij is an error term.
This simple additive linear function produces the main effects of our model. These
effects indicate how utility is affected by the level of the attribute when it is isolated from all
other attributes. Higher order effects indicate whether utility is also affected when two attributes
are presented in tandem (Louviere, et al 2000). We incorporate combinations of attributes by
interacting product attribute levels, so that,
ijijjjij cpxV εχπβ +++= (11)
Where cij is a vector of combination effects from product attribute level i's interacted with
product attribute level j's; and � is a conformable vector of coefficients to be estimated.
The final dimension to our model is that of preference variation among the population. In
order to account for different preferences among various sub-populations we incorporate
socioeconomic characteristics through interaction terms with the attribute level variables, so that,
ijijjjijij cpxsV εχπβα ++++= , (12)
where sij is a vector of socioeconomic variables i’s interacted with product attribute level j’s; and
� is a conformable vector of coefficients to be estimated.
16
Now that we have the utility function defined, we can model the choice as the relative
differences in utility. The difference between product A and product B for individual i is,
ABi
AB cpxsdV εχπβα +∆+∆+∆+∆= (13)
where dViAB = the utility difference between product A product B; �s = si *(xj
A - xjB); �x = (xj
A -
xjB); �p = (pj
A – piB); �c = (xj
A – xiA)*(xj
B – xiB); and �iAB = (ei
A- eiB) a normally distributed error
term.
The location attribute included four levels. We represented this as four binary variables:
� Unidentified = location (loc) 1 – loc 2, where loc is 1 if blank and 0 otherwise; � Local = loc
1- loc 2, where loc is 1 if “grown on a nearby farm,” 0 otherwise; � Ohio = loc 1 – loc 2, where
loc is 1 if “grown in Ohio,” or 0 otherwise; and � National = loc 1- loc 2, where loc is 1 if
“grown in U.S.,” 0 otherwise. Three of these binary variables were to be included in the model,
with the fourth excluded to prevent collinearity. However, statistical tests for the two groups
indicated that for both groups there was no significant difference in their evaluation of “grown
on a nearby farm” and “grown in Ohio”. Hence, these two locations were collapsed into a single
binary variable to indicate local production.
In addition to production location as a product attribute, the experiment included the
size/type of farm and a guarantee of freshness. Both of these attributes included only two levels.
Based on these attributes, two more binary variables are constructed: � Farm Size = size 1 – size
2, where size is 1 if “Fred’s Berry Farm,” 0 if “Berries Inc; � Fresh = fresh 1 – fresh 2, where
fresh is 1 if “Harvested Yesterday,” 0 if blank. Thus we have four x variables associated with the
vector in the first term: Local, Unidentified, Farm Size and Fresh. The price variable will be
included as �Price = price 1 – price 2, where the difference in price will range from $0.00 to
$2.00
17
Higher-order effects
Next we consider if there are any combinations of attributes that may produce significant
marginal effects. In the case of an unidentified product location, it is possible that another
attribute may negate the cue, as the signals may be substitutes. That is, if the strawberries have a
freshness guarantee, even though the location is unidentified, the consumer may be just as likely
to choose that product as if it had a local cue. The combination of a guarantee and unidentified
label, therefore, may have predictive value in the consumer choice beyond what each could
predict independently. The same might be true for the farm size, assuming that if it is produced
on a smaller scale, the consumer may assume that it will have a shorter supply chain. Likewise,
these cues may compensate for a national label as well. Preliminary models included several
interaction effects (Table 5).
Table 5 Higher order interaction effects for grocery sub-sample � National � Freshness � Unidentified � Farm Size � National -- � Freshness -- -- � Unidentified -- * -- � Farm Size * n/s n/s -- – indicates not tested, “n/s” if found not significant, and * if found significant at the 0.1 probability level
Interestingly, this same model showed no significant combination effects for the direct
market sub-sample. However, based on the significance of the coefficients for the grocery model,
we will include a combination of freshness with unidentified as well as farm size with national:
� Unidentified* � Freshness and � National* � Farm Size
Socioeconomic variables
Finally, we consider which socioeconomic variables will play a role in preference
variability. The socioeconomic variables must be interacted with each of the product variables
individually. Even if we only chose four of the most common demographic variables, we would
18
have 16 interaction terms (4 attribute variables * 4 socioeconomic variables). Thus, careful
selection of the appropriate interaction terms is required so that the model does not become too
large and lose too many degrees of freedom.
Table 6 shows the results from preliminary regressions for the direct market sub-sample.
Four interaction terms were found significant in this process, indicated by asterisks, two for
significance at 10% and three for significance at the 5% probability levels.
Table 6. Parameter estimates for demographic variables for direct market sub-sample Interaction Variable Price Local Unidentified Size Fresh Age -- n/s -- n/s ** Income n/s -- -- -- -- Education -- -- -- -- -- Gender -- ** -- -- n/s Metropolitan resident -- n/s -- * -- Local-support index -- n/s -- * -- Health index -- n/s -- -- -- Safety index -- n/s -- -- -- – indicates not tested, “n/s” if found not significant, and * (**) if found significant at the 0.1 (0.05) probability levels
The same process was completed for the grocery sub-sample, and interestingly, none of
the interaction terms were significant. For the sake of consistency, the same model will be used
for each of the sub-samples. Thus the socioeconomic variables included are, Gender* � Local:
where male =1, female = 0; Metro* � Farm Size: where city and suburban residents =1, rural =0;
Local Support* � Farm Size: where “main reason for buying local” is to support local business
or the regional economy =1, 0 otherwise; Age* � Freshness: where age is a continuous variable.
The model in equation 13 is estimated as:
Y = �1 – �2 � Price - �3 � Unidentified + �4 � Local + �5 � Fresh + �6 � Farm Size
+/- �7 � Unidentified* � Freshness +/- �8 � National* � Farm Size – �9 Gender* � Local
+/- �10 Metro* � Farm Size + �11 Local Support* � Farm Size +/- �12 Age* � Freshness
(14)
19
where the signs on parameters indicates the hypothesized relationship. The parameters are
estimated with the maximum likelihood procedure for a binary probit model with the Limdep
statistical package in NLOGIT version 3.0.
Before the model can be estimated, it is necessary to test the regularity of preferences
within the population, as well as within a group of experiments completed by one respondent.
Price sensitivity and attribute preferences will inevitably vary for each consumer; this variance
will be captured by the random error term. Because the model relies on a distribution of random
components which are independent and identically distributed, it is essential to extract any
systematic variance within the sample. A log-likelihood ratio test was used to test for differences
based on where the respondent was surveyed – direct market versus grocery store. Our results
suggested a systematic difference in response between these two groups, and thus we will
estimate separate models for these two groups. Similar tests were conducted for groups based on
household income, education and other socio-economic measures, but the log-likelihood test
statistic was not statistically different than zero at a 0.05 probability level.
In the same way that we tested for systematic variance in sub-populations, we also tested
for learning that occurs within the experiment. Because each respondent made eight consecutive
choices, it is possible that the observations are not independent of one another. If there is
systematic variance between the early and late experimental results, we can infer that learning
occurred during the experiments. We tested for this using a log-likelihood test for each of the
three sub-groups, a pool of the first three choices for each respondent, a pool of the last three and
a pool of the two in between. Again, the chi-square statistic was not sufficiently large to reject
the null hypothesis that learning occurred.
20
Table 7 shows the results of the final models. Note that each of the product attribute
parameter estimates were significant with the hypothesized signs except the “unidentified”
variable in the direct market model which was not significant. Comparison of the parameter
estimates across the models allows us to draw conclusions about the relative importance of these
cues for the two groups. For both groups, consumers value a freshness guarantee more than the
local cue. The grocery model shows a marginal effect of 0.31 for freshness compared to 0.23 for
local. In the direct market model, the marginal effect of freshness is 0.36 compared to 0.30 for
local. Likewise, both of these cues were valued more highly than the farm size cue and the
unidentified production location. For the grocery model, unidentified production location (-
0.116) had a larger effect than farm size (0.086), whereas the direct market model had a marginal
effect of 0.207 for farm size while unidentified was not significantly different than zero.
Table 7. Parameter estimates and marginal effects for the conjoint model. Grocery Direct Market
Product Specific Attribute Coefficient* Marginal effects Coefficient* Marginal effects
Intercept 0.053 0.021 0.067 0.027 � Price -0.901 -0.359 -0.646 -0.257 � Local 0.579 0.231 0.755 0.301 � Unidentified -0.292 -0.116 0.053 0.021 � Farm Size 0.217 0.086 0.520 0.207 � Freshness 0.785 0.313 0.893 0.356 Gender* � Local -0.072 -0.029 0.288 0.115 Metro* � Farm Size 0.006 0.002 -0.138 -0.055 Local Support* � Farm Size 0.056 0.022 -0.182 -0.073 Age* � Freshness -0.003 -0.001 -0.006 -0.002 � Unidentified* � Freshness -0.126 -0.050 0.045 0.018 � National* � Farm Size 0.187 0.074 0.139 0.055 * Estimates in bold refer to significance level of at least 0.10
We can also draw tentative conclusions about the underlying consumption services
associated with the cues. Assuming that the “harvested yesterday” guarantee is an effective way
21
of communicating freshness, we can factor out the freshness appeal from the local cue. The
marginal effects of local are significant holding freshness constant. That is, comparing two
products that are both harvested yesterday, consumers will still choose the local product more
often. Thus, we can view the effects as exclusive to the social component of the cue. If this
logic holds, both the grocery and the direct market models show substantial marginal effects for
the social component. In terms of the direct market model, the results suggest that freshness and
social ideology are both primary motivations to visit direct markets. Grocery store shoppers are
relatively more biased toward freshness as an attribute, but the social component is still strong.
The farm size cue “Fred’s Berry Farm” versus “Berries, Inc.” also shows a significant
coefficient, despite its potential for a wide range of interpretation. Our intended interpretation
was a “small-scale” versus “large-scale” production method. Other potential interpretations
include “corporate” versus “family-owned” operations, “national” production versus “local”, or
others. However, in general, we assumed that the cue, regardless of the exact interpretation,
would signal some form of social ideology revolving around supporting smaller, family-owned
business that runs parallel to the ideological component of the local cue. In the grocery model,
the marginal effect of the farm size cue is only 0.086, little more than a third as influential as
local. In the direct model, the farm size cue has an effect of 0.207, which more closely
approximates that of local. The large disparity in marginal effects for the local and farm size
cues for grocery could indicate two phenomenons. One explanation assumes that consumers
interpret the cues as communicating entirely different services. The result would suggest that the
service communicated by farm size is not as highly valued as that communicated by local. An
alternate explanation uses the same logic that assumes that freshness is factored out of the local
cue. If the two cues share some services in common but also have some exclusive to themselves,
22
we assume that those that overlap are factored out. Inferences about which services are shared as
opposed to distinct overextend the bounds of this analysis because more information is needed to
determine what exactly each cue communicates and how substitutable they are. We conclude
that farm size is not as strong a cue as local, whether it does not communicate as well, it
communicates a less valued service, or covers a narrower range of services.
The price variable in the grocery model shows a negative coefficient that is larger in
magnitude than any other product attributes. This indicates that grocery store customers are more
sensitive to changes in price than any other factor. The price impact is 38% greater than the local
impact and 14% more than freshness. In contrast, direct market consumers’ price sensitivity has
the third highest impact, less than both local and freshness. High price sensitivity is an
indication that the consumer is highly aware of the budget constraint and the nature of utility
gains as trade-offs. There are two plausible explanations for the discrepancy between these two
groups. A less favorable explanation in terms of the validity of our model is that people in the
direct market group are less aware of their budget constraints, so their responses may have a
higher degree of hypothetical bias. However, there is not a solid rationale for why direct market
consumers would be less aware of the trade-offs. More plausible is that direct market customers
are more willing to trade-off for the utility gains offered by the product attributes. Assume that
shopping at a direct market has additional opportunity costs, because it is in essence an extra trip,
since not all groceries can be purchased there. Those consumers who are willing to spend the
extra time are already demonstrating that they are willing to pay some premium for the products
and, thus, are willing trade off other utility gains for locally grown. Price sensitivity plays and
important factor in overall willingness to pay as will be reflected in our later calculations.
23
The expected sign for the “unidentified” production location variable shows up in the
grocery model, although it is not significant in the direct market. Assuming that the standard
strawberry that consumer’s purchase is identified as having been grown in the U.S., grocery store
consumers are almost 30% less likely to choose that product than one that does not identify
where it was produced. However, direct market consumers are no less likely to choose an
unidentified product than a national product. This result indicates that the uncertainty associated
with an unidentified product is not a major factor in their decision. Perhaps this stems from the
fact that direct market consumers are more trusting of products without information, since
typically, local produce is not packaged in this way. Alternately, this result could indicate that
“national” poses as much uncertainty in production methods as unidentified. Whatever the
reason, direct market consumers are more accepting of unidentified products than those at
grocery stores.
Significant variation in preferences was found in the direct market model among groups
based on gender and age. In contrast, none of these variables were significant in the grocery
model. When interacted with the local variable in the direct market model, the coefficient on the
gender variable indicates that males have stronger preferences for local than females. When age
is interacted with freshness, the coefficient indicates that older consumers value freshness less.
The original coefficient of .006 was evaluated at the mean age of 49 to produce the marginal
effect displayed in the table. All other demographic variables were insignificant, indicating a
high degree of preference regularity among consumers.
In contrast with the demographic variables, more combination effects were found
significant in the grocery model than in the direct market. For the interaction term between
unidentified and freshness, the marginal effect is negative. That is, there is less probability of
24
choosing a product whose label says nothing about location but has a freshness guarantee, ceteris
paribus. Also, the national cue combined with farm size shows a significant and positive
coefficient in both models. So, a label that says Grown in the U.S. on Fred’s Berry farm is more
frequently chosen.
By factoring out a consumer’s marginal utility of money, the coefficients in Table 7 can be
translated into willingness to pay in currency units. Each variable has a part-worth associated
with it, such that the attribute can be priced. Table 8 shows the estimates. When translated into
WTP measures, the disparity in consumer preferences among the models is even greater. This is
a result of the substantial difference in price sensitivity. The marginal utility of money is literally
the negative of the price coefficient. Notice that grocery consumers’ marginal utility of money is
over 25 percent higher than of direct consumers.
Recall that the national variable is the referent for the production location variable. These
WTP measures then, are calculated as part-worth values above a product that is grown in the
U.S. The average grocery store customer will be willing to pay an average of 64 cents extra for a
carton of strawberries if they are labeled “Grown in Ohio.” The estimate rises to $1.17 for the
average direct-market shopper.
Table 8 Comparison of grocery and direct WTP estimates
Grocery Direct
Marginal utility of money $0.90 $0.65
Local $0.64 $1.17
Unidentified -$0.32 $0.00*
Family farm $0.24 $0.80
Freshness guarantee $0.87 $1.38 * This coefficient was not statistically different from zero
For the production location variables then, our results show that consumers are willing to
pay a substantial premium for local strawberries. But the even larger premium is found for the
25
freshness guarantee cue. If the strawberries show a label that say harvested yesterday, consumers
will pay an average premium of between $0.87 and $1.38 for grocery store and direct market
customers, respectively. These premiums are considerable considering that the prompt indicated
that the typical market price for a carton of strawberries is only $3.00.
Farm size did not garner as substantial a premium as freshness or local. If the label on the
strawberries said Fred’s Berry Farm rather than Berries, Inc., consumers were willing to pay an
average premium of 24 cents for grocery store customers and 80 cents for direct market.
Conclusions
Our results suggest that consumers are willing to pay more for locally produced foods. In
the case of fresh strawberries, customers intercepted in grocery stores would pay an average of
64 cents more per quart, while those intercepted at direct markets would pay nearly $1.17 more
per carton of strawberries that was grown locally rather than berries identified simply as
"produced in the U.S." Our results also showed no significant difference in WTP for berries
“Grown nearby” relative to berries “Grown in Ohio.” This would suggest that to the median
consumer in both shopping venues, grown in Ohio is equivalent to local production.
These conclusions provide a solid rationale for the existence of niche market potential for
Ohio berry producers. Clear identification of Ohio-grown produce can distinguish a carton of
strawberries from another whose appearance is exactly the same. If local production is valued in
and of itself, consumers will realize higher utility if these products are available in the market
place at a comparable price. Likewise, producers and marketers who can document their
produce to be local can capture a portion of the consumer surplus as firm profit.
Another objective was to determine the extent to which the magnitudes of premiums vary
based on socioeconomic factors. Such knowledge would allow merchants to identify various
26
segments within the market and to advertise to or price for those segments strategically. For
instance, targeted discounting of prices based on the socio-economic characteristics allows
various local premia to exist simultaneously.
This study also proposed to determine the relative effectiveness of cues that communicate
the underlying consumption service of local. Freshness communicated through a guarantee
garners a higher premium, suggesting that it is more effective than local. The implication is that
labels that communicate the time of harvest are more relevant to a consumer’s decision than the
distance traveled from production location. Both types of information favor local producers,
since time is a function of distance in most cases. A local producer should decide which signal to
use, depending on the degree to which the distance factors into the time-frame from harvest to
retail. Direct markets have a definite advantage in this area. However, with advanced
coordination, retail outlets could garner premiums from the time-based freshness cue.
One further objective of this study was to decompose the appeal of locally-grown into its
freshness and home-bias consumption service. The conjoint analysis suggests that consumers are
primarily paying for the freshness of locally-grown produce. However, it does not explain all of
the appeal of locally-grown since nearly equivalent premiums still exist for local cues after the
freshness guarantee is controlled for. What we’ve variously called home-bias, social functions
and ideological components are legitimate utility-producing consumption services that come
close to rivaling that of freshness. The premiums associated with the family farm/farm size cues
suggest the viability of family farms or small-scale agriculture is another social function. While
we cannot specify the social function, in general we can assert that people are willing to use
consumption as a means to realize their social ideology. Thus, communicating a local region-of-
origin or small-scale production method is an important marketing strategy.
27
References
Aquino, Helen L.; Falk, Constance L. (2001), A Case Study in the Marketing of "Wolf Friendly"
Beef, Review of Agricultural Economics: 23, 524-37
Batte, Marvin T., Neal H. Hooker, Timothy C. Haab, and Jeremy Beaverson. " Putting Their
Money Where Their Mouths Are: Consumer Willingness to Pay for Multi-Ingredient,
Processed Organic Food Products." Food Policy. Forthcoming, 2006
Brown, Cheryl (2003), Consumers' preferences for locally produced food: A study in southeast
Missouri, American Journal of Alternative Agriculture: 18, 213-224.
Buchardi, Henrike, Carsten Schröder, and Holger D. Thiele (2005) Willingness-To-Pay for Food
of the Own Region: Empirical Estimates from Hypothetical and Incentive Compatible
Settings, Selected Paper prepared for presentation at the American Agricultural
Economics Association Annual Meeting, Providence, Rhode Island, July 24-27.
Campbell, Robert, G. Nelson, Benjamin L., Ebel, Robert C., Dozier, William A., Adrian, John L
and Hockema, Brandon R. (2004), Fruit Quality Characteristics that affect Consumer
Preferences for Satsuma Mandarins, HortScience: 39(7), 1664-1669.
Dimitri, Carolyn and Catherine Greene. 2002. “Recent Growth Patterns in the U.S. Organic
Foods Industry.” U.S. Department of Agriculture, Economic Research Service.
Agriculture Information Bulletin Number 777. 2002.
Ganiere, Pierre, (2004) Impact of consumer behavior on the willingness to pay for genetically
modified foods, M.S. Thesis, The Ohio State University.
Govindasamy, R., J. Italia. 1999. “Predicting Willingness-to-Pay a Premium for Organically
Grown Fresh Produce.” Journal of Food Distribution Research July (1999): pp. 44-53.
28
Hanemann, Michael, and John Loomis (1991), Statistical Efficiency of Double-bounded
dichotomous choice contingent valuation, American Journal of Agricultural
Economics:73, 1255-63
Huffman, Wallace E. (2003), “Consumers' Acceptance of (and Resistance to) Genetically
Modified Foods in High-Income Countries: Effects of Labels and Information in an
Uncertain Environment.” American Journal of Agricultural Economics, 85: 1112.
Hurley, S.P., and J.B. Kliebenstein (1998) “Market potential for environmental improvements in
livestock production – who will pay?” Presented at the International Conference on
Animal Production Systems and the Environment, Des Moines, Iowa, July 21.
James, Jennifer S. James, Fleischer, Shelby, Johnson, David H., Schwab, Bradley D. and Lord,
John (2005), Consumer Purchases of Biotech Sweet Corn: Results from a Market
Experiment, Selected paper prepared for presentation at the American Agricultural
Economics Association Annual Meeting, Providence, Rhode Island, July 24-27, 2005.
Johnston, R.J., C.R. Wessells, H. Donath and F. Asche. 2001. “A Contingent Choice Analysis of
Ecolabeled Seafood: Comparing Consumer Preferences in the United States and
Norway,” Journal of Agricultural and Resource Economics, 26(1):20-39.
Lancaster, K. (1966), A New Approach to Consumer Theory, Journal of Political Economy,
74:132-157.
Loomis, John, Brown, Thomas, Lucero, Beatrice, Peterson, George (1997), Evaluating the
Validity of the Dichotomous Choice Question Format in Contingent Valuation,
Environmental and Resource Economics: 10(2): 109-123.
29
Loureiro, Maria L. and Susan Hine. 2002 A Comparison of Consumer Willingness to Pay for A
Local (Colorado-Grown), Organic, and GMO-free Product. Journal of Agricultural &
Applied Economics, vol. 34 iss. 3: pp. 477-487.
Loureiro, Maria Luz; McCluskey, Jill J. (2000), Assessing Consumer Response to Protected
Geographical Identification Labeling, Agribusiness: 16, 309-20.
Loureiro, Maria L., Jill J. McCluskey and Ron C. Mittelhammer, (2002) Will consumers pay a
premium for eco-labeled apples? Journal of Consumer Affairs: 36, 203-17.
Loureiro M.L. and W.J. Umberger (2004), A Choice Experiment Model For Beef Attributes:
What Consumer Preferences Tell Us.” Selected Paper Presented at the American
Agricultural Economics Association Annual Meetings, Denver, Colorado, August 1-4.
Louviere, J.J., D.A. Hensher, and J.D. Swait (2000), Stated Choice Methods: Analysis and
Application. Cambridge: Cambridge University Press.
Lusk, Jayson L.; Hudson, Darren (2004), Willingness-to-Pay Estimates and Their Relevance to
Agribusiness Decision Making, Review of Agricultural Economics: 26, 152-69.
Mabiso, Arthur, James Sterns, Lisa House and Allen Wysocki (2005), Consumers’ Willingness-
To-Pay for Country-Of-Origin Labels in Fresh Apples and Tomatoes: A Double-Hurdle
Probit Analysis of American Data Using Factor Scores, Selected Paper prepared for
presentation at the American Agricultural Economics Association Annual Meeting,
Providence, Rhode Island, July 24-27.
Onyango, Benjamin (2004), An Application of Choice modeling To Measure U.S. Consumer
Preferences for Genetically Modified Foods. Paper prepared for presentation at the
American Agricultural Economics Association Annual Meeting, Denver, Colorado,
August 1-4. (Scarpa 2005).
30
Teisl, M.F., B. Roe and R.L. Hicks. 2002. “Can Eco-labels Tune a Market? Evidence from
Dolphin-Safe Labeling,” Journal of Environmental Economics and Management 43(3):
339-359
Thompson, G.D. and J. Kidwell. 1998. “Explaining the Choice of Organic Produce: Cosmetic
Defects, Prices, and Consumer Preferences.” American Journal of Agriculture
Economics vol. 80, iss. 2 (May 1998): pp. 277-87.
Umberger, Wendy J.; et al. (2002), U.S. Consumer Preference and Willingness-to-Pay for
Domestic Corn-Fed Beef versus International Grass-Fed Beef Measured through an
Experimental Auction, Agribusiness: 18, 491-504.Underhill, Sheila E., and Enrique E.
Figueroa. 1996. Consumer Preferences for Non-Conventionally Grown Produce. Journal
of Food Distribution Research vol. 27: pp. 56-66.
van der Lans, Ivo A. (2001), The Role of the Region of Origin and EU Certificates of Origin in
Comsumer Evaluation of Food Products. European Review of Agricultural Economics,
28: 451.
Wang, Q. and Sun, J. 2003. “Consumer preference and demand for organic food: Evidence
from a Vermont survey.” Paper prepared for American Agricultural Economics
Association Annual Meeting July (2003): pp. 1-12Wessells et al.1999
Williams, Pamela R. D., and James K. Hammit. 2000. A Comparison of Organic and
Conventional Fresh Produce Buyers in the Boston Area. Risk Analysis vol. 20 iss. 5: pp.
735-746
Williams, Pamela R. D., and James K. Hammit. 2001. Perceived Risks of Conventional and
Organic Produce: Pesticides, Pathogens, and Natural Toxins. Risk Analysis vol. 21, iss. 2:
pp. 319-330
top related